Urban residential clustering and mobility of ethnic groups: impact of fertility

The study addresses how residential clustering and mobility of ethnic and linguistic groups in an urban context relate to fertility and intraurban migration. In the context of increasing urbanization, fertility decline, and growing multiethnic populations, understanding residential segregation and its demographic drivers is critical. The authors propose quantifying segregation and population dynamics using scaling-law-based frameworks built on null models of spatial distributions and flows, extending concepts widely used in urban science. They focus empirically on Greater Helsinki (Finland), where high-quality register microdata enable fine-grained spatial and temporal analysis across socioethnic groups. The purpose is to (1) measure spatial clustering, (2) identify socioeconomic and geographic correlates (income, distance to centre, socioeconomic similarity), (3) model intraurban migration flows, and (4) connect clustering with mobility and fertility to explain how patterns develop and persist.

The paper situates itself within several paradigms of segregation research: spatial indices (capturing multiple dimensions of segregation), cartographic visualization, dynamic modelling, and inclusion of demographic factors. Prior work has focused largely on the US and major European regions with longer immigration histories, generally treating segregation as societally undesirable. Urban scaling, fractality, and power laws are established in urban systems for macro quantities and temporal evolution; recent work examines intra- and inter-urban migration dynamics within these frameworks. In Finland, segregation of ethnic minorities has been widely studied thanks to register data, including descriptive analyses, segregation indices, integration efforts, and policy impacts (housing, labor markets). The Helsinki region is notable for strong official desegregation policies and high concentration of residents with migrant backgrounds, making it an appropriate setting for this analysis.

Data and setting:

• Anonymized Statistics Finland register microdata with yearly modules on demographics, locations, and migration, 1987–2020.
• Study area: Greater Helsinki (17 municipalities); residential locations mapped to 1 km × 1 km grid cells; 4,088 populated cells on average.
• Spatial clustering primarily analyzed for year 2019; migration analyzed from 1995 onward with flows aggregated in 5-year windows (t in 1995, 2000, 2005, 2010, 2015; Δ=5).
• Population in 2019: 1.57 million in Greater Helsinki; ~233k with foreign background; ~191k assigned a unique country of ethnicity; plus ~83k Swedish-speaking Finns.

Ethnicity assignment and grouping:

• For non-Finnish-background individuals born in Finland: country of ethnicity from citizenship at migration (preferring first citizenship).
• For those born abroad: country of ethnicity from first immigration event to Finland.
• Individuals with multiple migrations (ambiguous COE) were excluded.
• Groups follow modified GBD 2017 super-regions: Central Europe–Central Asia; East Asia; High-income countries; Latin America; North Africa–Middle East; sub-Saharan Africa; South East Asia–Pacific Islands; South Asia; plus Russia, Somalia, Baltic countries, Scandinavia; and Swedish-speaking Finns.

Spatial clustering model:

• Start from a null (unsegregated) distribution of group populations across area units.
• Base model (per group): regress log of group count in cell i on total group size and local neighbourhood group presence (Moore neighbourhood of 8 adjacent cells), capturing spatial clustering via an exponent coefficient.
• Full model adds covariates that may independently influence clustering:
  • Mean disposable income at location (proxy for housing price and sorting).
  • Distance from city centre (centralization dimension).
  • Similarity in socioeconomic status between group g and local population measured by KL divergence (lower distance = higher similarity).
• Estimation: Ordinary Least Squares (OLS) per group using 2019 data; variance inflation factor (VIF) checks (generally <5); acknowledge complexity in estimating the clustering coefficient due to simultaneity.
• Robustness: Spatial Autoregressive (SAR) lag model reformulation with spatial lags of log group counts in neighbouring cells; estimated using R spatialreg; high agreement with OLS coefficients (Pearson correlations ~0.77–0.99 across parameters). Hierarchical linear regression across groups compared null, base, and full models; adjusted-R^2 improved at each stage.
• Filtering: require group count n_ig < total n_i and n_ig ≥ N_min, with N_min=10; use log n_ig := 0 if below threshold; similar threshold for migration models.

Migration models (intra-urban):

• Compare local inflows and outflows for groups; intra-regional moves dominate local changes relative to international and inter-regional flows.
• Null model for flows across area units; extended base models:
  • Inflow model: include local presence of group (n_ig) and neighbourhood outflow of the group (F_out from adjacent cells) to capture social gravity/local relocation propensity; population-level attractiveness via total flows.
  • Outflow model: attributes of focal unit (e.g., group presence).
• Full migration models add distance to centre and mean income at location.
• Flows aggregated over 5-year windows; variables mean-centred within each time window; VIF <5. Ten of 13 minority groups had inflow model adjusted-R^2 between 0.4 and 0.8; outflow models generally had higher explanatory power.
• Combine inflow and outflow to analyze inflow/outflow ratio and differences in coefficients (Δα terms) for population, income, and centralization effects.

Fertility measure:

• Child–woman ratio (CWR) per group in 2019: number of children under 5 divided by number of women aged 15–49; used as a cross-sectional fertility proxy to relate to spatial clustering and migration coefficients.

• Spatial clustering levels:
  • Clustering coefficient (from full model) ranged approximately 0 to 0.5 across groups; highest for Swedish-speaking Finns; generally lower than base-model estimates after adding covariates (notably reduced for High-income countries and Somalia).
  • Native Finnish-speaking population showed negligible clustering.
• Clustering–fertility link:
  • In base models, clustering increased with child–woman ratio (CWR). Excluding Swedish-speaking Finns and High-income countries as outliers, slope ≈ 0.35 (p<0.05, R=0.43).
• Effects of covariates in spatial model (full):
  • Location attractiveness (total population at location): β_N > 0 for most groups; strongest for North Africa–Middle East origins; weakest for Swedish-speaking Finns.
  • Mean income at location: β_m ≤ 0, indicating inverse relation between group population and local mean income; pronounced for some groups (e.g., Somalia), weak or absent for High-income countries, Scandinavia, Latin America, Swedish-speaking Finns.
  • Distance from centre: generally inverse relation (β_cen < 0), but often not significant; strongest decrease with distance for High-income countries.
  • Socioeconomic similarity (KL): positive association (β_KL > 0) indicating tendency to co-locate with similar socioeconomic profiles.
• Migration patterns and models:
  • Intra-regional migration flows dominated local population change for minorities.
  • Inflow models: adjusted-R^2 between 0.4–0.8 for 10/13 minority groups; outflow models had slightly higher explanatory power; VIF <5.
  • Clustering–mobility connection: In base models, clustering (β) strongly associated with neighbourhood movement coefficient α (slope 1.54, p<0.001, R^2=0.74), excluding Swedish-speaking Finns outlier. Jointly, β varied with both CWR and α in base models (adj. R^2=0.88; F2,8=39.04; p<0.001). In full models, β related significantly only to α (adj. R^2=0.73; F2,9=16.16; p<0.01).
  • Density dependence of inflow/outflow ratio: Δα_a = α_a^(in) − α_a^(out) < 0 for groups, implying areas with higher local group concentration tend to see relatively greater outflows over time; Δα_a positively associated with CWR (slope 0.26, p<0.01, R^2=0.18). Correlations of CWR with α^(in), α^(out), and Δα_a were 0.28, −0.05, and 0.44, respectively.
  • Income gradient in migration: Relationship between spatial β_m and migration Δα_m ≈ unit slope (slope 1.04, p<0.05, R^2=0.32), indicating consistent link between spatial income association and inflow/outflow differences by income.
  • Centralization in migration: β_cen vs Δα_cen positive slope (0.64) but not statistically significant (p>0.1, R^2=0.18).
• Group-specific insights:
  • Swedish-speaking Finns: highest spatial clustering and highest socioeconomic similarity; clustering not well explained by migration or fertility, likely reflecting long-term historical concentration and additional cultural/institutional factors.
  • High-income countries: clustering mainly associated with neighbourhood migration; these populations can maintain residence in advantageous neighbourhoods, reinforcing concentration via local moves.

The modelling framework clarifies how multiple factors jointly shape residential clustering of socioethnic groups. First, it identifies separate roles of population attractiveness, income, centrality, and socioeconomic similarity in predicting group distributions. Second, it empirically connects spatial clustering exponents with mobility parameters: short-distance (neighbourhood) migration tendencies strongly relate to higher clustering, indicating that local moves reinforce spatial concentration. Third, fertility contributes to clustering in geometric (base) models, consistent with larger families co-residing and elevating local densities; however, in full models this fertility effect is absorbed by income differences, suggesting economic context mediates fertility–clustering associations. Fourth, across groups there is a general tendency to move away from high-concentration areas, with the rate negatively correlated with fertility—together implying diffusion-like dynamics across adjacent areas. These findings align with prior Helsinki research using other segregation measures, which observed dilution of immigrant neighbourhoods via mobility and increased concentration via higher fertility. The framework adds interpretability by linking spatial and flow exponents and by quantifying how income and centrality gradients in migration map onto spatial patterns. Outlier behaviour of Swedish-speaking Finns underscores how historically embedded institutions and long-standing concentrations can produce stronger clustering beyond what fertility and local mobility alone explain.

The study introduces a scaling-law-based framework that integrates spatial and migration null models with socioeconomic covariates to explain residential clustering of ethnic and linguistic groups in Greater Helsinki. It shows that spatial clustering is reinforced by local (short-distance) migration and, in geometric models, is positively associated with fertility; income and centrality further shape patterns, and socioeconomic similarity contributes to co-location. The strong linkage between spatial clustering parameters and migration coefficients provides a coherent interpretation of how intraurban movement sustains or diffuses concentrations. Future research should test these relationships in other urban contexts with larger and more diverse migrant populations, investigate temporal dynamics with finer-grained flow networks (origin–destination pairs), and incorporate additional demographic and institutional variables (e.g., housing market constraints, school catchment policies) to assess generalizability and mechanisms.

• Limited number of observations for certain analyses and groups; some regression fits had lower explanatory power due to smaller sample sizes (e.g., Scandinavians).
• Spatial clustering assessed primarily cross-sectionally (2019), while migration spans earlier years; simultaneity addressed but residual endogeneity may persist.
• Base and full models do not directly control for age, sex, and education; assumed their effects are captured by income and socioeconomic status; potential omitted-variable bias remains, particularly regarding neighbourhood group presence.
• Assumption of no omitted variables significantly correlated with neighbourhood group presence (log ñ_ig) may be violated.
• Flow models use aggregated inflows/outflows by focal locations rather than full origin–destination pairs to mitigate sparsity/noise, potentially limiting gravity dynamics precision.
• Filtering thresholds (N_min=10) and data availability constraints (suppression for areas with <3 residents) may affect estimates.
• Special cases (e.g., Swedish-speaking Finns) indicate that historical and institutional factors not explicitly modelled can strongly influence clustering.